2015 IRS-TPC Research Conference
Improving Tax Administration
Through Research-Driven Efficiencies
June 18, 2015
2015 IRS-TPC Research Conference
Welcome
Eric Toder
Institute Fellow: Urban Institute, and
Co-Director: Urban-Brookings Tax Policy
Center
Alain Dubois
Deputy Director: IRS Office of Research,
Analysis, and Statistics
John A. Koskinen
Commissioner of Internal Revenue
2015 IRS-TPC Research Conference
Session 1: Innovative Methods for Improving Resource Allocation
Moderator:
Melissa Vigil
IRS, RAS, Office of Research
Estimating Marginal Revenue/Cost Curves
for Correspondence Audits
Alan Plumley
IRS, RAS, Office of Research
Examining the TDA Collectability Curve
Jeff Wilson
IRS, Taxpayer Advocate Service
Analysis of Flow-Through Entities Using
Social Network Analysis Techniques
Shannon Chen
University of Texas at Austin
Discussant:
Arnie Greenland
University of Maryland
Estimating and Using
Marginal Revenue/Cost Functions
for Correspondence Audits
Team:
Ron Hodge, Kyle Richison, and Getaneh Yismaw (RAS:R)
Nicole Misek, Matt Olson, and Sanith Wijesinghe (MITRE)
Alan Plumley
18 June 2015
Government Accountability Office:
“IRS Could Significantly Increase Revenues by Better Targeting Enforcement Resources”
(December 2012)
5
ESTIMATING MARGINAL REVENUE/COST
Government Accountability Office:
Cumulative Revenue ($)
“A hypothetical shift of a small share of resources (about $124 million) from exams of tax returns in less productive
groups… to exams in the more productive groups could have increased direct revenue by $1 billion… (as long as
the average ratio of direct revenue to cost for each category of returns did not change).”
Slope = Marginal
Revenue/Cost at
Point A
A
(e.g., last year’s
overall result)
Slope = Average
Revenue/Cost at
Point A
Cumulative Cost ($)
6
ESTIMATING MARGINAL REVENUE/COST
• The average R/C is constant only if
we select cases randomly.
• If we’re successful in giving priority to
the most cost-effective cases, then
the average should decline as we
increase the budget.
• Making allocations based on average
R/C overstates the impact of a
change in budget.
Is There Evidence of Marginality?
That is, are the methods used to prioritize potential audit
inventory better than random selection?
• If we had to reduce the number audited, would we tend to avoid the
least cost-effective ones?
• If we were able to increase the number audited, would the additional
ones be less cost-effective than the ones already selected?
Answer: YES, there is evidence of marginality!
7
ESTIMATING MARGINAL REVENUE/COST
Current Research Project
Focused on correspondence (rather than face-to-face) audits
• Many categories of correspondence audits (“projects”), defined by what
tax return line item is being questioned
• Begin by trying to improve resource allocation across the projects
assuming no change to operational selection system
8
ESTIMATING MARGINAL REVENUE/COST
Method (for a given project and tax year)
Step 1: Identify for each closed audit:
 The tax revenue collected from the audit;
 An estimate of the cost to identify, assess, and collect that revenue (from
time applied, series, grade, etc.); and
 The value of the variable used operationally to assign priority to the audit.
Step 2: Sort all closed audits in declining order of the prioritization
variable.
When a return was audited is generally not a proxy for priority.
Step 3: Derive for each audit the cumulative revenue and cumulative
cost of all returns audited having same or greater priority.
Step 4: Plot cumulative revenue vs. cumulative cost.
9
ESTIMATING MARGINAL REVENUE/COST
Correspondence Audit Project Code A2, Tax Year 2006
The curvature away from the average
line is a measure of marginality—the
power of the prioritization method to
identify returns with higher
revenue/cost (at the left).
10
ESTIMATING MARGINAL REVENUE/COST
Correspondence Audit Project Code A2, Tax Year 2006
Step 5: Fit a curve through the plot. A
simple quadratic curve displayed an
excellent fit for most projects.
11
ESTIMATING MARGINAL REVENUE/COST
Cumulative Revenue ($)
Deriving Marginal Revenue/Cost Functions
Slope = Marginal
Revenue/Cost at
Point A
A
slope of the cumulative curve at each point)
• If the cumulative plot is a
quadratic curve, then marginal
R/C declines linearly as a
function of cumulative cost.
Returns sorted in declining order of priority
Cumulative Cost ($)
12
Step 6: Mathematically derive the
marginal Revenue/Cost curve (the
ESTIMATING MARGINAL REVENUE/COST
Marginal Revenue/Cost as a Function of Budget, Tax Year 2006
• As more budget is applied to a project, the
marginal R/C declines.
• Greater curvature in the Cumulative Revenue vs.
Cumulative Cost plot corresponds to steeper
slope in the Marginal R/C plot.
• How should the available budget be allocated to
these projects to maximize direct revenue?
13
ESTIMATING MARGINAL REVENUE/COST
Marginal Revenue/Cost as a Function of Budget, Tax Year 2006
• Direct revenue is maximized when the marginal
R/C ratio is the same for all projects.
• Otherwise, revenue could be increased by
shifting resources from projects with low MR/C
to those with higher MR/C.
14
ESTIMATING MARGINAL REVENUE/COST
Actual vs. Optimal* Revenue and Cost ($ M)
Actual Allocation
Project
Total
Revenue
Total
Cost
Optimal* Allocation
Net
Average Marginal
Total
Revenue Rev/Cost Rev/Cost Revenue
Total
Cost
Change Change
in
in
Net
Average Marginal
Cost
Revenue
Revenue Rev/Cost Rev/Cost
C1
$27.6
$4.0
$23.6
6.90
6.81
$47.7
$7.0
$40.8
$6.87
6.51
$2.9
$20.1
A1-Lo
$151.9
$13.4
$138.5
11.33
9.98
$293.3
$30.5
$262.7
$9.60
6.51
$17.1
$141.4
A1-Hi
$18.6
$2.4
$16.2
7.83
4.29
$14.2
$1.6
$12.6
$8.84
6.51
-$0.8
-$4.3
A2
$61.7
$7.1
$54.6
8.67
5.16
$53.3
$5.7
$47.6
$9.30
6.51
-$1.4
-$8.4
A3
$21.9
$4.3
$17.6
5.12
4.82
$1.6
$0.2
$1.4
$7.24
6.51
-$4.1
-$20.3
A4
$36.3
$3.0
$33.4
12.22
9.83
$52.7
$4.9
$47.8
$10.71
6.51
$1.9
$16.3
O
$287.1
$37.0
$250.0
7.75
5.41
$205.7
$21.2
$184.5
$9.70
6.51
-$15.8
-$81.4
Total
$605.1
$71.2
$533.9
8.50
$668.6
$71.2
$597.4
$9.39
6.51
$0.0
$63.5
* These are “optimal” allocations only in the sense that they maximize net direct revenue in the
absence of any constraints. They are not truly optimal since they don’t account for indirect
effects, other benefits or costs, constraints, or valuations of benefits and costs other than dollars.
15
ESTIMATING MARGINAL REVENUE/COST
Next Steps
 Work with Campus Exam and OCA to incorporate MR/MC
functions in the Exam Planning Scenario Tool to support the
FY16 Exam Plan.
 Identify constraints
 How much can we expand projects with high MR/MC?
 How easily can examiners work different projects?
 Make assumptions about indirect effects
 Minimum coverage constraints, etc.
 Improve workload selection (screening, prioritization variables)
16
ESTIMATING MARGINAL REVENUE/COST
IRS Collectability Curve
Taxpayer Advocate Service
Research & Analysis
June 2015
Focus
• Taxpayer Delinquency Account (TDA) liabilities.
• Individual Master File
• Collection statute (generally 10 years)
Background
• Over 50 percent of the IRS Individual Master File (IMF) TDA inventory
has been in the function assigned the delinquency for at least 10
months.
• Over 70 percent of the IMF TDAs in IRS inventory at the end of 2014
are Tax Year 2010 and prior liabilities.
• Over 20 percent of the IMF TDAs have less than four years remaining
on the collection statute, meaning that the delinquency has existed
for over six years.
Objectives
• Determine dollars collected during each year after TDA assignment.
• Distinguish between TDA dollars collected from subsequent payments
and offsets.
• Determine how dollars collected vary by categories of TDA balance
due.
Objectives
• Determine how TDA dollars collected vary by type of assessment (selfreported and IRS imposed)
• Quantify the effect of assessed penalties and interest on the TDA
balance due.
• Determine the percentage of TDA liabilities abated by the IRS and if
abatements vary by sources of assessment.
Objectives
• Determine the percent of TDA cases full paid within 10 years.
• Determine if TDA dollars collected vary by collection channel.
Methodology
• Analyzed balance due (from IMF ARDI) at initial TDA assignment.
• Analyzed cases entering TDA status (by year entered) from calendar
year 2003 through 2012.
• Used IMF data to distinguish between subsequent payments, offsets,
penalties, interest, and adjustments (classified by transaction code).
Methodology
• Used ARDI major source of assessment to determine type of IRS
assessment (self-assessed or IRS imposed).
• Used TRCAT code to determine if case was assigned to ACS, collection
queue, or CFf.
Limitations
• Changes in module balances include assessed and accrued penalties
and interest; however, the specific finding on penalties and interest
only includes assessed amounts.
• Amounts abated because of accepted offers in compromise are
included in the sections on abatements; however, in FY 2014,
accepted offers in compromise only accounted for about one percent
of the initial TDA balance.
Findings
• Dollars collected generally decreased by more than 50 percent from
the first year to the second year.
• In the third year, collections decrease by about a third from the
amount collected in the second year.
Findings – Percent of Payments Collected per
Year
Percent Collected by Years Elapsed
60%
50%
40%
2003
2005
30%
2007
2009
20%
2011
10%
0%
1
2
3
4
5
6
7
8
9
10
Years Elapsed
Findings - Dollars collected also decline as a
percent of the available balance.
Findings – Dollars Collected by Subsequent
Payment Have Decreased
• In the earlier years, dollars collected by subsequent payment are
nearly triple dollars collected through offset however, in more recent
years, this margin has decreased to only double.
• Overall, dollars collected on TDAs by subsequent payment appear to
be decreasing from 2003 to 2012.
Findings – Percent of Dollars Collected vary by
TDA Balance
• Nearly Three-quarters of TDAs have balances of $5,000 or less.
• However, over 80 percent of the balance owed is contained in TDAs
with balances above $5,000.
• The IRS Collects a Higher Percentage of dollars when the TDA Balance
is Smaller.
Findings – Percent of Dollars Collected vary by
TDA Balance
• The IRS offsets the highest percent of dollars to TDAs under $5,000.
• As time progresses, the percent of TDAs with initial balances of
$5,000 or less are decreasing, while the percent of the total initial
TDA balance above $5,000 is increasing.
Findings – Source of Assessment
• On a percentage basis …
• The IRS Collects twice as much from TDAs resulting from self-assessments
than from audit assessments.
• In recent years, the IRS also collects about twice as much from TDAs
resulting from self-assessments than AUR assessments (the difference was
not quite as large in earlier years).
Findings – Source of Assessment
• The IRS generally collects the least on SFR assessments.
• The IRS collects the highest percentage from offsets on TDAs from
AUR assessments.
• The IRS generally offsets a slightly higher percentage on TDAs from
audit assessments than self-reported assessments.
Findings – Source of Assessment
Self-Reported
Assessments
Substitute for
Return
Audit
Assessments
AUR
Assessments
Trust Fund
Recovery
Penalties
16%
2003
18%
4%
12%
34%
6%
31%
17%
2005
20%
5%
20%
32%
6%
24%
25%
12%
2007
20%
5%
25%
36%
6%
9%
21%
24%
9%
2009
15%
4%
20%
28%
6%
7%
15%
21%
8%
2011
10%
2%
12%
25%
4%
Audit
Assessments
33%
Substitute for
Return
Year
Year
Self-Reported
Assessments
Trust Fund
Recovery
Penalties
Offsets
AUR
Assessments
Subsequent Payments
2003
56%
14%
23%
2005
60%
13%
28%
2007
51%
10%
2009
45%
2011
40%
Findings – Assessed Penalties
and Interest
• Over equivalent three year periods, assessed penalties and interest have remained
relatively constant; however, when sufficient time has elapsed, penalties and
interest are increasing.
Percent of Liability (actually due) Attributable to Penalties and
Interest
30%
25%
20%
15%
Cumulative %
10%
% after 3 Years
5%
0%
2003
2005
2007
2009
2011
Findings - Abatements
• Generally, the IRS abates from a quarter to a third of TDA
assessments.
Year
Initial TDA Balance
$ Abated
% Abated
2003
$15,326,191,192
$2,985,977,270
19%
2005
$25,996,084,845
$8,066,761,341
31%
2007
$40,678,451,308
$13,086,103,480
32%
2009
$41,987,700,518
$10,716,623,485
26%
2011
$42,926,217,917
$11,990,870,525
28%
Findings - Abatements
• SFR assessments have the highest abatement rate (nearly 50 percent in some
years).
• Abatements attributable to AUR assessments are growing, while self-reported
assessments are the least likely to be abated.
Findings – Abatements by Source of
Assessment
Year
Self-Reported
Assessments
Substitute
for Return
Audit
Assessments
AUR
Assessments
Trust Fund
Recovery
Penalties
2003
6%
49%
15%
15%
39%
2005
6%
47%
12%
29%
40%
2007
12%
43%
14%
28%
35%
2009
9%
36%
13%
27%
28%
2011
16%
40%
19%
18%
29%
Findings – Full Payments and Collection Channel
• The full payment rate has been decreasing for TDAs issued in more
recent years, while a higher percentage of the initial TDA liability
remains due.
• ACS collects a higher percent of TDA dollars by both subsequent
payment and offset than CFf (may be a reflection of inventory
composition).
• Often, over a third of TDA dollars assigned to CFf are abated.
Conclusions
• Dollars collected in aggregate and as a percent of the balance due decrease
significantly during the first three years after the IRS assigns a liability to TDA
status.
• When continuing to look at the collection of liabilities after the third year of the
initial TDA assignment, collections continue to dwindle and the reduction in the
module balance declines almost completely.
• Overall, dollars collected through the offsets of other overpayments are much less
than dollars collected through subsequent payments.
Conclusions
• Delinquent modules with balances due not in excess of $5,000 comprise the
vast majority of TDAs. However, over 80 percent of the total amount due
resides with TDAs with balances greater than $5,000.
• The IRS collects both a higher percentage of subsequent payments and offsets
in the lowest balance due categories.
• The percent of the TDA balance collected is significantly greater for selfreported liabilities than when the IRS makes additional assessments.
• Penalty and interest significantly increase the balance owed by taxpayers,
particularly when the underlying balance remains unresolved for several
years.
Conclusions
• The IRS abates between a quarter and a third of TDA liabilities. The IRS
abates about 40 to 50 percent of its substitute for return (SFR) assessments.
• The IRS completely resolves most of its TDA modules within the 10 year
collection statute, with a resolution rate of about 80 percent for TDAs
assigned in 2003 and 2005. Unfortunately, the percent of TDAs resolved
appears to be declining for TDAs initiated in later years. Additionally, the
balance owed on these delinquencies has only been reduced by less than 50
percent.
• ACS realizes the largest percent of TDA balances collected by subsequent
payment and offset. While the percent of dollars abated is high in all TDA
collection channels, the abatement rates are significantly higher in the queue
and CFf than in ACS.
ANALYSIS OF FLOW-THROUGH ENTITIES
USING SOCIAL NETWORK ANALYSIS
TECHNIQUES
Ashish Agarwal
Shannon Chen
The University of Texas
at Austin
Rahul Tikekar
Ririko Horvath
Larry May
IRS, RAS
Advisory Roles: Robert Hanneman ( UC Riverside), Lillian Mills (UT Austin)
Research Question(s)
• Application: Can Social Network Analysis (SNA) be a
useful technique for IRS “big data” analysis of flowthrough entities?
• Compliance Risk: Do the ways “enterprises” embed
flow-throughs in their corporate structure facilitate
noncompliance?
• Do SNA characteristics of greater network complexity explain tax
noncompliance?
• (How) Do loss flow-through entities create more compliance risk?
Prior Evidence
• Prior work examines the association between firm characteristics
and corporate noncompliance.
• Mills (1998) finds a positive association between book-tax difference and proposed IRS
audit adjustments.
• Hanlon, Mills, & Slemrod (2005) examine firm size, industry, multinationality, public vs.
private firms, choice of executive compensation, and corporate governance.
• Some academic work on complexity and tax avoidance or tax
risk generally.
• Wagener and Watrin (2013) find that organizational complexity (number of subsidiaries,
ownership chain length, cross-country links, and ownership percentage) is associated with
greater income shifting incentives.
• Balakrishnan et al. (2012) argue that tax avoidance increases financial complexity as
evidenced by decreased corporate transparency.
Prior Evidence (cont.)
• Some academic work on choice of overall business structure.
• e.g., Guenther (1992), Ayers et al. (1996), Gordon & MacKie-Mason (1994),
MacKie-Mason & Gordon (1997)
• Some recent academic work on use of special purpose entities,
which include LLCs, LLPs, trusts, and other flow-through entities.
• Feng et al. (2009) & Demere et al. (2015)
• However, there is a lack of empirical evidence on the effect of
flow-through entities on tax noncompliance specifically.
Data Sample
Year
Number of Enterprises
Entities
k-1 links
Parent-Sub links
Primary-Secondary links
Sample Based on
Proposed Deficiency
Database
2009
5,913
107,638
411,644
75,832
55
• The following pictures describe SNA variables.
Random
Sample
2009
5,000
31,884
28,210
1,225
2,590
Sample Enterprise Plots
Enterprise X
Enterprise Y
Preliminary Evidence on Our Research
Questions
• Effort last summer yielded learning how to use YK1 data and
applying SNA approach to measure various nodal and linkage
characteristics of about 6,000 enterprises in the 1120 LB&I
taxpayer population for 2009.
• Some measures of network complexity are associated with higher
detected noncompliance (proposed deficiencies).
• Controlling for raw predictors of audit selection like size, profitability, DAS.
SNA Measures
Network Measure
Density
Diversity
Definition
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
1
𝑛𝑛 𝑛𝑛 − 1
2
� p2𝑖𝑖
𝑖𝑖
Degree Centrality
External Degree Centrality
Closeness Centrality
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 𝑝𝑝𝑝𝑝𝑝𝑝 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁
𝑛𝑛 − 1
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
A𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
∑𝑛𝑛𝑗𝑗=1 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝑖𝑖, 𝑗𝑗
𝑛𝑛 − 1
−1
1.0
0.2
0.5
Node Diversity
0.05 0.10 0.20
0.1
0.01 0.02
Node Diversity
0.50 1.00
Node Diversity
2
5
10
20
50
200
500
Number of Nodes
PDD Sample
2000
6
8
10
12
Number of Nodes
Random Sample
14
5
4
2
3
Average Degree Centrality
6
5
4
3
1
2
1
Average Degree Centrality
Degree Centrality
2
5
10
20
50
100
200
Number of Nodes
PDD Sample
500
2000
2
4
6
8
10
Number of Nodes
Random Sample
12
16
Centralization & Node Level Degree Centrality
Centralization = 0.05
Centralization = 0.1875
Centralization = 0.45
2.0
1.8
1.6
1.2
1.4
External Degree Centrality
2.0
1.8
1.6
1.4
1.0
1.2
1.0
External Degree Centrality
External Degree Centrality
2
5
10
20
50
100
200
500
2000
2
4
Number of Nodes
PDD Sample
6
8
10
Number of Nodes
Random Sample
12
16
0.500
0.200
0.050
0.005
1e-06
0.020
1e-04
1e-02
Closeness Centrality
1e+00
Closeness Centrality
2
5
10
20
50
100
200
500
2000
2
4
8
10
Number of Nodes
Number of Nodes
PDD Sample
6
Random Sample
12
16
2.0
1.8
1.6
1.4
1.2
1.0
External Degree Centrality
Outlier Analysis
1
5
10
50
500
Number of Nodes
5000
Identifying Economically Important Nodes
Relationship Between Deficiency and SNA Measures
(Preliminary Analysis)
• Regression:
Deficiency = a0 + a1 Assets + a2 DAS + a3 NetIncome +
a4 ClosenessCentrality + a5 Nodes + a6 Degree +
a7 NodeDiversity + a8 DegreeCentrality
• As expected, Deficiencies are higher for larger and more profitable firms.
• Relevant to our question, Deficiencies are significantly higher when the
nodes are further away (a4<0) or when the node type is more
concentrated (a7<<0).
Project Status
• Initial contract for Ashish Agarwal and Shannon Chen ended
September 2014, simple results shown today.
• Waiting to re-establish IPA and Disclosure.
• Great opportunities for future work when access restored.
Future Work
• Refine measures
• Generate measures for multiple years
• Conduct validation of measures
• Explore other enterprise definitions
• Contribute to tax administration of complex organizations
• Academic Paper on Noncompliance (Agarwal, Chen & Mills)
2015 IRS-TPC Research Conference
Session 1: Innovative Methods for Improving Resource Allocation
Moderator:
Melissa Vigil
IRS, RAS, Office of Research
Estimating Marginal Revenue/Cost Curves
for Correspondence Audits
Alan Plumley
IRS, RAS, Office of Research
Examining the TDA Collectability Curve
Jeff Wilson
IRS, Taxpayer Advocate Service
Analysis of Flow-Through Entities Using
Social Network Analysis Techniques
Shannon Chen
University of Texas at Austin
Discussant:
Arnie Greenland
University of Maryland